Taking your definition of an abstract model (so we don't squabble over mere definitions), I don't think that just by removing information you'll go from an actual baseball to the 'abstract concept' of a sphere. You'll also be adding information. For example, for your model you can provide the formula that will yield the exact volume of the sphere - you can't do that as precisely for your baseball. Will your abstract models typically be more compact / contain less information than your baseball, sure. However, the information may be partially different, not just a subset, which it would be if you were just ignoring information.

That's true. Balls are very complex, so there isn't actually much you can ignore about them without invalidating your results. But you can ignore a lot of things and get approximately correct results, which is usually good enough when talking about balls.

Numbers, however, tend to be a little more convenient. If there's a hole in the bag of apples which you don't take into account, you'll get bad results, because that's a detail which impacts the numeric aspect of the apples. But we don't really care that it's a hole when talking about the number of apples. All we need to keep in mind is that the number decreased. If 1 apple fell through the hole, you can abstract that to a simple -1.

Anyway, this post has gotten out of hand, mostly because I was unclear, so I'll retract it and use these comments to write a hopefully clearer version. Thanks for the feedback.